So I think we've heard so far today that AI can be very helpful in addressing and tackling very narrow tasks. And you wanna really think about where it may make sense for you, your practice and your patients. My name's Ivy Lee, I get the privilege of working with all of these amazing panelists on the Augmented Intelligence Committee. And I'm so excited today to share with you how I use artificial intelligence for quality improvement projects in my practice and for quality of life. And my hope is really to inspire you to try to experiment and play with these tools and figure out where it may be helpful. And we're also gonna play a game of find the hallucination. So take a look at this image. Where do you see the hallucination? I see a hand raised, yeah. The hands, exactly. Gen AI has such a hard time with hands. And so you can see that there's a couple of distal digits that are missing. So I have no financial disclosures to disclose. And we're not gonna delve into the robust science of quality improvement. There's a lovely CME in the JAD from last year about that. But we are gonna talk about kind of the broad domains of quality improvement as well as talk about the impact of quality of life. And really think about how this may be helpful in terms of tools within your medical toolkit. So for context, I practice concierge medicine. I have a boutique membership-based model in Los Angeles. And so my practice prides itself on accessibility to highly personalized care. And the culture of my practice is really to continuously learn and adapt my practice to support the needs of my patients and to personally expand the capacity of my team and myself so that we love what we do and we don't burn out. So that's kind of the North Star in which I orient myself to. This is Dr. Atul Butte. He is the chief data scientist for the whole UC system. And he has coined this wonderful term called scalable privilege. And kind of references AI's potential to scale the personalized experience of medicine. And that's what my hope is in terms of using some of these tools is how can I scale this type of quality of care and very highly personalized care to the patients that I see in person to also my telehealth patients who are not necessarily affluent or located closely to my physical office. Here's an example of an image that I generated on DALI. Hope you can find the hallucinations. And this refers to kind of, again, focusing on our needs and our pain points as a way to think about where to start on your AI journey. So where's the hallucination here? Exactly, there's a snake stethoscope. What's that, right? Like fascinating in terms of generation of this image here. So where I start in my office is I think about people, processes, and the purpose. And about once a month we have a session where we think about what are the most common complaints or what was the pain point among my staff and I. And then from that we think about, we upvote or downvote how relevant it is to my team. Is this a problem that's felt by one person? Is this a problem that's felt by multiple people? And then think about then from their different perspectives how we come up with an aim statement. So this is just on a little posty note in our lunch area. And then we come up with SMART goals, things that are specific, measurable, achievable, relevant and time bound so that we can really hold ourselves to this narrow kind of scope and goal. And then as we think about that, we think about how does this impact our workflow? What are the processes involved? How can we improve that? Is this a task or a problem where we can solve with AI? Oftentimes it's not, but sometimes it is. And when we do think about that, we go back to that process that Dr. Rodenberg had talked about, what do we need to tackle? And what is the data that we need? Do we have that accessible? Is this gonna have to change our workflow in terms of information gathering? How do we look at algorithms? Can we use something that's already off the shelf and easily accessible? Do we need to buy an outside tool? And for a small practice like mine, we're trying not to buy all these one-off solutions, but we want something that's really easy and it's affordable and something that is not difficult to train people to use. So then after that, we think if AI is a solution, we pick one tool, we pick a certain amount of time in which we will pilot this, and then we get feedback from our team and from our patients. And throughout this process, we're keeping it very transparent in terms of my patients of this is what we're trying to do, to trying out to make sure that we deliver a very personalized and high quality experience for you. So I like to start out with a really low risk and high impact tasks. That's where my risk tolerance is for. And I really want to triage those different needs and those different pilots to what I'm most comfortable with. And so looking at this list of tasks that Dr. Schlesinger had just talked about in terms of large language models, these are tasks that I know that I could potentially outsource, right? Or think about using AI tools for. And I really wanna use these tools to cut the scut and really think about those tasks that are peripheral to why I went to medical school, peripheral to clinical decision making, because I actually really like that cognitive challenge and the friction involved in terms of decision making. And thinking about this then, I have to then ask myself, what is a tool that's gonna align with my goals, right? My medical values of first to no harm, my moral values, my epistemic values of how much explainability do I need? What's my threshold for trusting this tool? And then our societal values of transparency and then also equity. And after I go through this personal process of validation, then I can choose things and tools that align with my values, my risk tolerance, my patient preferences, and kind of then define what trustworthy means to me as I think about integrating these tools. This is a wonderful book. It includes excerpts from Dr. Koh and Dr. Novoa, so a fun beach read. And then in terms of those tasks that are low risk and high impact for me, they really center in three categories. One is communication, two is creatives, and three is care. And you'll find that all of these kind of relate to the type of comprehensive and really concierge care that I provide. And in terms of communication, it really helps in terms of tailoring content to my patients as well as the providers that I work with in telehealth and within my referral system, as well as public when I'm engaging in public messaging and making sure that what I share for my patients is very culturally appropriate. I use these tools, the large language models, a lot to brainstorm and think through how I have hard conversations or how I pitch different topics. And oftentimes, as Dr. Rodenberger said, I'm not looking to make things longer or more fluffy language or multisyllabic. I'm looking at, okay, this is content that I'm thinking about disseminating or sharing in an interview or sharing in print. Please summarize the key facts. It's kind of my teach back, right? Am I effective in my prose in making sure that those key messages are actually the ones that people are gonna walk away with? For the media, in terms of doing research on things that I'm not aware of, so between trend and skincare, right? No idea. It was great background research to prepare me for an interview for that. And then to also speak to that audience of saying, you know, what resonates with this demographic, with this language preference? And helps me also create some visuals in terms of making sure that people get that core message. In terms of telehealth, again, tailoring education, collapsing and summarizing guidelines, which is really easy, and then attaching the reference sources can be helpful for our partners in delivering care. So I think that this has been hugely helpful in terms of delivering more culturally competent care and offering that more personalized experience. One of the tools that Dr. Schlesinger referenced is one that I use a lot for answer finding, it's a large language model that then refers to the reference sources. So it's great in that it can help me find information quickly and pull out those reference sources that then I can share with patients. And this is a free service. Elicit is also very helpful. And both of these tools, Perplexity and Elicit, use retrieval augmented generation, so RAG, to kind of hone and improve the accuracy in terms of their generated results. And I would say that this has supplanted the PubMed searches in my practice. This is also great if you wanna do meta-analyses and review articles because it can summarize your data and reference sources pretty easily. The second category that I use AI in my practice is to generate creatives. This is a shameless plug for our fun social event that we are hosting tomorrow. And this took me about 10 seconds. Like two or three iterations, 10 seconds, using CREA and then editing in Canva. Who can spot the hallucination here? Yes, sir. Does that tree floating on the cloud? Yes, that totally is nonsensical, right? What else? There's another hallucination. Any volleyball players here? The volleyball seams are off, right? So that's not actual volleyball. It looks like one, but the seams are off. So, again, really fun to create this design. This is not gonna be my side hustle, like Dr. Nouveau, who's a budding graphic artist, but gives me the tools to do something that makes it easy, super fun, and very time efficient. Okay, Chuck Close is one of my favorite artists. He does these lovely collages of his self-portraits, and this is so embarrassingly mine. And I have no patience to go sit and do a photo shoot, but I knew that I had to do a couple of different images for different speaking engagements. One was to a biker group on skincare, right? So you can see up top, there is biker chick me, right? Who can spot the hallucination there? I have two right hands, a broken wrist as well, right? Because Gen AI has a really hard time generating hands. There's one Norman Rockwell version of me, the middle school picture of me with the blue background. I've got one with what I call the stoner version of me with really blown out pupils on the bottom. And then the top left there, I have severe strabismus. And you always have to be careful with these, because that is now my profile pic within my family, so super embarrassing and will live on for a while. So great for a website, socials. And I use the tools that I use, I usually use OpenAI's chat GPT-4, because it has Dolly integration. And you can see the bottom right, sorry, bottom left is my slide opening photo, but you can see how many iterations it took to get there. And again, all of this is probably within the time span of 15 minutes, right? So when I first started off the prompt and my prompts are in the handouts, are that we start out with like a utopic vision of a doctor looking into the future with innovation. You can see some of the biases that are already there. I actually specified female physician, but I didn't specify ethnicity. So the default ethnicity is white. And then you can see variations, and you can use this again to iterate and to tailor this to your need and what you're actually looking for. And again, very fun to experiment and play with and refine with your own human feedback and your preferences. So in terms of care, again, referencing how we use these tools for delivering care. Again, I can really specify my research in looking in mining the medical literature. I can tailor letters. I don't enter any personal information into using these AI tools, because remember, HIPAA often, unless it's integrated within your HR, this is not under the purview of HIPAA. And then I use it for rehearsing hard conversations or tailoring education, working with the community and working with advocacy groups. There's one fun pilot that's going on right now in my office. We're using visual prompting for a computer vision model to look at laser and procedure tray setups. Is this a way that I can train and help my staff to make sure that when I enter the room, my setup is needed for the specific laser and for that specific patient. Again, trying to improve efficiency in the times that I'm in the office. And in terms of telehealth and media as well. So again, helping me provide more culturally competent care. And I'm hoping to work on creating my patient-facing bot that is trained on patient education materials that I've created. So you can see in this kind of short talk, we've covered those main domains of quality improvement. And this has dramatically improved the experience for my patients and the quality of life for my team and myself. So you may say, Ivy, where do I start? This is just released about a week ago. It's the AMA's report on future of health. The AAD was represented in terms of contributing our perspectives on where AI can help and support our role in an augmented way. And you can see adoption of AI and its current use cases is still relatively low. So I would really encourage you to tap into your AI FOMO and experiment and play and assess where your greatest unmet needs are, where these tools may help you and add value to your practice and your patients. And that's a great place to start. Again, thinking about low risk, high impact. And then in terms of this journey, think about how you can play and tinker and co-evolve with this technology. And I really invite you to reimagine our work and our mission and to really tap into the human intelligence, your human intelligence, the human neural network that is here as we think about how we elevate that patient experience and really extend our healthcare system beyond sick care and towards individualized goals of health and wellness. Because I do think these tools can really expand our capacity. And really through experimenting with this, we can also participate in the design of these. And that participatory design, I think together we can build better products because we understand our patients and our needs the best. And I think that's the way that we really build tools that are worthy of our trust and worthy of our patients' trust. So, thank you. Thank you.

artificial intelligence

healthcare

quality improvement

patient care

personalized approach